You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Is your feature request related to a problem? Please describe
ClusterManager uses the committed ClusterState on node for serving Read Requests. If the applied cluster-state is not in-sync with cluster-manager, nodes fetche the entire cluster-state from cluster-manager. In large clusters, appliers might take significant time to commit the cluster-state.
During this duration, the read requests on the node fetch the cluster-state from cluster-manager as the term-version will not match.
Describe the solution you'd like
Cluster Manager has two phases :- Publish and Commit to propagate the Cluster State updates from leader to follower.
As part of publish phase, the updated cluster-state is published to all nodes in cluster. The nodes cache the published cluster-state locally. Transport handlers of Read Admin API's fetch the cluster-wide committed term and version from cluster-manager. Nodes can utilize cached cluster-state that is published but not yet committed locally if the term-version matches.
This will help especially for large clusters where there is a significant delay in order of seconds (~18s in 1000 node cluster) between the publication and commit of cluster-state.
Related component
Cluster Manager
Describe alternatives you've considered
No response
Additional context
No response
The text was updated successfully, but these errors were encountered:
Is your feature request related to a problem? Please describe
ClusterManager uses the committed ClusterState on node for serving Read Requests. If the applied cluster-state is not in-sync with cluster-manager, nodes fetche the entire cluster-state from cluster-manager. In large clusters, appliers might take significant time to commit the cluster-state.
During this duration, the read requests on the node fetch the cluster-state from cluster-manager as the term-version will not match.
Describe the solution you'd like
Cluster Manager has two phases :- Publish and Commit to propagate the Cluster State updates from leader to follower.
As part of publish phase, the updated cluster-state is published to all nodes in cluster. The nodes cache the published cluster-state locally. Transport handlers of Read Admin API's fetch the cluster-wide committed term and version from cluster-manager. Nodes can utilize cached cluster-state that is published but not yet committed locally if the term-version matches.
This will help especially for large clusters where there is a significant delay in order of seconds (~18s in 1000 node cluster) between the publication and commit of cluster-state.
Related component
Cluster Manager
Describe alternatives you've considered
No response
Additional context
No response
The text was updated successfully, but these errors were encountered: